AbstentionBench - underspecified context - SQuAD 2.0 - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.1 70B | 99.39 |
| 2 | Gemini 1.5 Pro | 98.5 |
| 3 | DeepSeek R1 Distill Llama 70B | 98.39 |
| 4 | GPT-4o | 98.37 |
| 5 | Llama-3.1-Tulu-3-8B-DPO | 97.82 |
| 6 | Llama 3.1 8B | 97.78 |
| 7 | Llama-3.1-Tulu-3-8B | 97.74 |
| 8 | Llama 3.1 70B Instruct | 97.57 |
| 9 | Llama 3.3 70B Instruct | 97.4 |
| 10 | O1 | 97.3 |
| 11 | Llama-3.1-Tulu-3-8B-SFT | 97.28 |
| 12 | Llama 3.1 405B Instruct | 97.09 |
| 13 | Llama-3.1-Tulu-3-70B | 96.96 |
| 14 | Llama 3.1 8B Instruct | 96.62 |
| 15 | Llama-3.1-Tulu-3-70B-DPO | 96.43 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-squad-2-0-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.